my lora only work with anime or stylized models. realistic models is not compatible with this lora.
A and B is trained on different dataset, while I used the same training settings.
a lot of the images, I have manually tagged to get better result.
if you wonder why the lora is so large that is because I used 1000 dalle3 images that need way bigger lora than normally and I think this is my best dalle3 lora.
I recommend you use masterpiece, best quality, amazing quality, realistic, and add 3d, anime, 2d, bad quality, worst quality, worst detail, sketch, censor, simple background, in negative.
check out my pixiv if you are interested in my work. https://www.pixiv.net/en/users/34413372